Skip to content

quitgenius/ai-engineering

Repository files navigation

AI Engineering Toolkit

Production-tested Claude Code skills, commands, and tools for AI-augmented software development.

Based on learnings from AI-Augmented Development at Scale.

Structure

ai-engineering/
├── claude-skills/      # Skills (complex, stateful workflows)
│   ├── spec/           # /spec - Interactive spec creation
│   ├── beads/          # /beads - Task decomposition with git-backed tracking
│   ├── converge/       # /converge - Exhaustive multi-pass research
│   ├── sparky/         # Sparky integration (Letta persistent memory)
│   └── aws-cdk-development/
│
├── claude-commands/    # Commands (slash commands for Claude Code)
│   ├── create_plan.md  # /create_plan - Architecture planning
│   ├── implement_plan.md
│   ├── research_codebase.md
│   └── ... (25+ commands)
│
├── claude-agents/      # Custom subagent definitions
│   ├── codebase-analyzer.md
│   ├── codebase-pattern-finder.md
│   └── ...
│
└── sparky-rs/          # Rust Slack bot for Sparky (Letta agent)
    └── src/

Core Workflow

/spec → /create_plan → /beads → /implement_plan → Visual Verification
  ↓         ↓            ↓           ↓                    ↓
Human     Human        Human      Claude              Claude
Scope     Guide       Verify     Execute            Verify UI

Key Skills

/spec - Scoping

Interactive spec creation with:

  • Parallel research (codebase, Linear, Sparky, Git, docs, Exa)
  • Strategic questions based on research findings
  • Gherkin acceptance criteria
  • Automatic Linear ticket creation

/create_plan - Architecture

  • ADR compliance checking
  • Structured decision questions
  • Phased implementation planning
  • File:line references from research

/beads - Task Decomposition

Uses beads for:

  • Atomic, context-window-sized tasks
  • Git-backed distributed tracking
  • Dependency graphs for parallel execution
  • Quality gates per task

/converge - Deep Research

Multi-pass research until convergence:

  • Normal (~60 min): 3 parallel agents → sequential refinement
  • Extreme (~3 hours): + GPT-5 Pro synthesis
  • Insane (~6+ hours): GPT-5 Pro handles everything

Installation

Skills & Commands

Copy to your Claude Code configuration:

# Copy skills
cp -R claude-skills/* ~/.claude/skills/

# Copy commands
cp -R claude-commands/* ~/.claude/commands/

# Copy agents (optional)
cp -R claude-agents/* ~/.claude/agents/

Sparky (Letta Agent + Slack Bot)

See sparky-rs/README.md for setup instructions.

Requires:

  • Letta account and agent
  • Slack workspace with bot permissions

Dependencies

License

MIT

About

Pelago Health AI Engineering Repository

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

No releases published

Packages

 
 
 

Contributors